Robot will race in city

Jun Xian Leong Jan 29, 2007

Move over, Sandstorm and H1ghlander, Carnegie Mellon’s robots that participated in a mock military mission through the Mojave Desert in 2005. There’s a new car on the block, and it has more power than either of them.

Meet Boss, Tartan Racing’s latest entry to the Defense Advanced Research Projects Agency’s (DARPA) Grand Challenge. DARPA’s Grand Challenge is an annual competition, set this year for November 3, which pits self-driving vehicles against each other on a course that simulates military supply missions in a variety of environments.

Most students on campus have probably heard of Carnegie Mellon’s Red Team and its exploits in the Grand Challenge over the past years. The team this year, however, was intended to be broader-based than the Red Team, involving more faculty than before. Byron Spice, co-director of media relations for the School of Computer Science, said that the name Tartan Racing was chosen to differentiate this year’s team from the Red Team, even though it is still being led by professor William “Red” Whittaker and includes some veterans of the Red Team.

Whereas the previous two Grand Challenges have been held in the desert, this year’s Grand Challenge ups the ante by moving the course to a city setting — hence rendering the name DARPA Urban Challenge. The differences between this year’s challenge and previous challenges are significant. In particular, the harsh winds and treacherous dunes in the desert course have been replaced with traffic and a maze of confusing roadways in the urban course.

Electrical and computer engineering major Ziv Wolkowicki, who is currently working on developing software for the robot’s sensors, said, “This race is significantly different.... Vehicles have to do things that they didn’t have to do before.” The robot, for instance, will need to yield to other contenders using its sensory and perceptual devices, he said.

Sandstorm and H1ghlander, Red Team’s previous entries into the competition, were optimized for operation in desert and rocky terrain, with strong chassis and rugged, weatherproof builds. However, Boss will be dealing with far less harsh environments, smoother terrain, and an overall less-punishing course, allowing for a lighter build. At the same time, Boss will face challenges that arise from racing in an urban setting.

Previous contenders in the DARPA Grand Challenge have had to navigate around obstacles within a largely open environment and arrive at the finish line within a specified time limit. Urban Challengers, however, must also follow traffic roads and rules, severely limiting their freedom of movement and navigation. To make things more difficult, Boss will also need to deal with moving traffic, traffic circles, and busy intersections without remote assistance and still arrive at the finishing point on time.

Whereas previous Grand Challenges were more focused on testing the stamina and endurance of the robots, this year’s competition places more emphasis on intelligent reaction to a changing environment and computation of decisions in real-time. Accordingly, the new vehicle carries more sensors and computing power than the previous racers. Intelligence, not speed, is the essential factor for winning this race.

The self-driving 320-horsepower Chevrolet Tahoe has 25 sensors, three miles of extra wiring, 10 computers with 208,000 lines of new computer code, 64 off-the-shelf components, and 350 custom-built parts, plus the car’s standard equipment. Its guidance system includes radar, laser sensors, a GPS system, and motion cameras to accurately decipher its location and detect nearby vehicles.

The choice of car reflects the team’s primary sponsor, General Motors. “That’s why it’s a Chevy,” Spice said. “Why a Tahoe? Lots of room in back to accommodate computers, plenty of height to give sensors a good view, enough horsepower to produce the electrical power to run the computers, sensors, and the high-powered AC that keeps it all cool.”

Boss performed admirably in its first test run, which was held last Tuesday. The vehicle successfully navigated the test course on the LTV site at Hazelwood at speeds up to 20 miles per hour, allowing user-driven vehicles to pass if they arrived first and gracefully handling intersections.

“We run all sorts of tests,” Wolkowicki said, including long-range tests to assess the robot’s level of endurance and intersection tests to make sure the robot yields to other vehicles when necessary.

DARPA will give $2 million, $1 million, and $500,000 awards to the top three finishers that complete the course within the six-hour time limit. However, Whittaker said that Tartan Racing’s main aim is not merely to win the reward, but, “more importantly, to make driving safer, to create new autonomous navigation and robotic technologies, and to change the world’s perception of what is possible.”

Wolkowicki said that the team has been progressing quickly. He said that many people are involved in different aspects of the project, ranging from marketing to software develoment. He said, “People are very devoted.”